Originally posted by: dbaena

Hello,

I have a problem with this instruction in cplex 12 with C++. I think I know the problem but I don't sure. I have a problem that I solve and I obtain:

Warning: Control callbacks may disable some MIP features.
MIP emphasis: optimality.
MIP search method: traditional branch-and-cut.
Parallel mode: none, using 1 thread.
 
Node log . . .
Best integer =   5.232939e+05  Node =       0  Best node =   5.189600e+04

I do:

double BestUB= cplex.getObjValue();

after if i do cout << BestUB I obtain the value: 523294 and not 5.232939e+05 ¿ok?

After, I do:

cta->cplex.setParam(IloCplex::CutUp,BestUB - Eps); where Eps = 0.00001.

I solve the problem I obtain: 523294

I think that I have a numerical problem, but How can I do it...

Thanks a lot!
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

> dbaena wrote:

>
> Node log . . .
> Best integer = 5.232939e+05 Node = 0 Best node = 5.189600e+04
>
> {code}
>
> I do:
>
> double BestUB= cplex.getObjValue();
>
> after if i do cout << BestUB I obtain the value: 523294 and not 5.232939e+05 ¿ok?

That's probably due to the default formatting of the output stream. You can put a formatter in the cout line to get more accuracy if you wish.
>
> After, I do:
>
> cta->cplex.setParam(IloCplex::CutUp,BestUB - Eps); where Eps = 0.00001.
>
> I solve the problem I obtain: 523294

If you're expecting a strictly smaller value, the first thing you need to do is print getObjValue to more decimal places. Even then I'm not sure your epsilon is big enough to do the job -- it's changing the 11th significant (decimal) digit, which puts it out where I think rounding error could plausibly overwhelm it.

Paul

Mathematicians are like Frenchmen: whenever you say something to them, they translate it into their own language, and at once it is something entirely different. (Goethe)
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: dbaena

Thank you Paul,

but i don't understand the behaviour of CutUp Cplex instruction...

I do:

cta->cplex.setParam(IloCplex::CutUp,UB - Eps);

where UB - Eps = 8203.35922

and I solve the problem with cplex.solve().

I obtain a solution with value:

8203.35923

and it is not normal I think...
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Dani,

I'm not positive, but I think rounding error may be at play here. CutUp is defined in terms of the original problem. I think that it is enforced in terms of the preprocessed problem. So perhaps it is possible that the solution is less than or equal to the cutoff in the preprocessed problem, but not when it is converted back to the original problem.

I'm also not sure which of CPLEX's various "epsilons" might be used in making the decision (EpOpt? EpRHS? presumably not EpInt). The default for EpOpt is 1e-6. If we allow for some rounding in the reported objective value, maybe it's really within 1e-6 of your CutUp value?

Paul

Mathematicians are like Frenchmen: whenever you say something to them, they translate it into their own language, and at once it is something entirely different. (Goethe)
#CPLEXOptimizers
#DecisionOptimization